OpenAI to acquire Ona, expanding Codex with persistent cloud

OpenAI will acquire Ona, the startup that provides secure, persistent cloud environments for code execution, the companies announced this week. The purchase is explicitly aimed at integrating Ona's runtime technology into Codex to support long-running AI agents and persistent state across enterprise workflows.

OpenAI described the deal as a way to give Codex the ability to host durable execution contexts that survive beyond single API calls, enabling agents to maintain state, coordinate multi-step tasks and run background processes inside controlled cloud sandboxes.

What OpenAI is buying

Ona builds managed runtime instances designed to run code securely over extended periods. Those environments include features tailored for production usage: persistent storage, isolated networking, resource metering, and lifecycle controls for background processes. The product targets developers who need remote execution that remains available between sessions, unlike ephemeral serverless containers that terminate after a single request.

Under the terms described, Ona's engineering team, runtime platform and underlying orchestration layers will be folded into OpenAI's developer tooling group. Financial terms were not disclosed. OpenAI said the acquisition will prioritize compatibility with existing Codex APIs, and that customers will be able to opt into Ona-backed runtimes as a managed service for production agents and workflow automation.

How Codex will use Ona

The integration targets three technical gaps in current large language model deployments. First, persistence: Codex calls today execute statelessly, which complicates multi-step automations. Ona's environments provide durable state and storage that agents can reference across hours or days. Second, secure execution: Ona's sandboxes enforce network and file-system constraints, which helps enterprises meet compliance and data-governance requirements when models need to act on internal systems. Third, orchestration: Ona supplies lifecycle controls and resource quotas so teams can run background jobs, schedule tasks, or scale agents without managing bespoke infrastructure.

OpenAI positions the combined stack for use cases such as continuous monitoring agents that maintain a memory of prior interactions, automated code maintenance bots that run tests and deploy changes over time, and cross-system workflows that require sustained access to databases or APIs. The company says it will surface these capabilities through Codex SDKs and a managed console that lets teams control runtimes and permissions.

The deal also signals an answer to a practical developer question: where does model reasoning stop and executable infrastructure begin. By owning the runtime layer, OpenAI can offer an end-to-end proposition that pairs model outputs with an approved execution environment, rather than leaving that integration entirely to customers.

Why it matters

Bringing Ona into Codex shifts more of the execution stack under OpenAI's control, reducing friction for teams that want models to take persistent actions. Enterprises that need secure, auditable long-running agents will gain a turnkey option, while competing cloud and developer-tool vendors may face pressure to match integrated runtimes. The move tightens the link between model capabilities and operational infrastructure, changing how production AI workflows are assembled and managed.